On June 29, 2026, an anonymous X account named @heatedrivalryai published a 25-page dossier that ignited a firestorm of outrage, harassment, and deleted stories across the Heated Rivalry fanfiction community. At the center of the controversy: a single CSS class buried in the source code of fanworks hosted on Archive of Our Own (AO3). The account claimed that the presence of font-claude-response-body proved that 38 popular stories were generated, at least in part, by Anthropic’s Claude AI—and provided a browser-style tool to make that marker blindingly obvious. Within days, the dossier was being called the “fan fiction Panama Papers,” and the hunt for AI-generated prose had turned vicious.

The uproar wasn’t another abstract debate over whether something “sounds” like AI. It was built on a concrete, inspectable artifact: a formatting remnant left behind when text is copied directly from Claude into AO3’s rich-text editor. That narrow, technical clue quickly became a community-wide verdict—and a case study in how easily a forensic trace can be mistaken for a conviction.

The Tiny HTML Clue That Set Off a Firestorm

The class font-claude-response-body is applied by Anthropic’s web interface to AI-generated responses under specific conditions. When a user selects text in Claude and pastes it directly into AO3’s editor, that HTML markup can survive, embedded invisibly in the page’s source. The @heatedrivalryai account exploited this by creating a public AO3 site skin—a custom CSS theme that anyone can apply to their reading experience. The skin scans a work’s HTML for the string font-claude-response-body. If it finds it, the entire page turns an alarming shade of red.

The Verge independently tested the mechanism and confirmed that it works as described: text pasted straight from Claude triggered the red screen, while identical text routed through Word, Google Docs, or even a plain-text editor did not. That makes the marker more reliable than probabilistic AI “detectors” like Turnitin’s or GPTZero, which guess at authorship based on statistical patterns and often produce false positives. But it also makes the marker vastly less comprehensive than many fans assumed.

The site skin does not detect any other AI model—not ChatGPT, not Gemini, not Llama. It cannot flag Claude-generated text that was copied into any other editor along the way. It provides no information about how much of a story was written by AI, whether the author merely brainstormed ideas or ran grammar checks through the model, or whether the text was heavily rewritten before being published. In short, the red screen proves one thing only: that a specific copy‑and‑paste path was used at some point.

What the Marker Means for Readers—and What It Doesn’t

For readers scanning the Heated Rivalry tag, the site skin became a quick-and-dirty filter. A story that turned red was instantly suspect; a story that didn’t was presumed clean. That binary logic spread rapidly across social media, where screenshots of red pages were shared as evidence of deception. But the leap from “a known string is present in the HTML” to “this author is a fraud” collapses several important distinctions.

  • The marker’s absence does not mean a work is AI‑free. A writer could have used Claude extensively, pasted everything into Notepad first, and the skin would never alert.
  • The marker’s presence does not mean the work is entirely—or even mostly—AI‑generated. A single paragraph pasted straight from the tool could trigger the flag.
  • The marker does not reveal whether the writer intended to mislead. AO3 itself encourages the use of optional tags such as “Created Using Generative AI,” and many authors use AI for legitimate assistance: translation, grammar correction, overcoming writer’s block.

In the hands of a community already primed to see AI use as a betrayal, the red screen functioned less like a disclosure and more like an accusation engine. It collapsed a complex spectrum of AI involvement into a green‑light/red‑light signal, and the social penalties for triggering that red light were swift and severe.

For Writers: The Red Screen Is a Warning, but Also a Blunt Instrument

If you’re a fanfic author, the existence of the font-claude-response-body marker should prompt a moment of careful reflection—not panic. Here’s what the episode means for your own workflow and presence on AO3.

If you use Claude for any kind of drafting assistance: Be aware that pasting directly from Claude into AO3’s editor can leave a visible trail. If you wish to avoid the marker, the simplest workaround is to paste the text into a plain‑text editor (Notepad, TextEdit, or even the URL bar) first, then copy it again before pasting into AO3. This strips the legacy HTML and prevents the red screen from activating. However, stripping the marker in this way is purely a technical step; it does not resolve the ethical question of whether you should disclose your AI use.

If you want to be transparent: AO3 offers the freeform tags “Created Using Generative AI” or “AI‑Generated Text.” These tags can help set reader expectations, but the Heated Rivalry backlash shows why many writers hesitate. Once a work is tagged as AI‑involved, some readers may treat it with immediate hostility, regardless of how the tool was actually used. Even minor admissions can spiral into accusations that an entire back catalog is fraudulent.

If you’re worried about false accusations: There is no perfect defense. A user could apply the site skin to your work and share a screenshot to a platform designed for pile-ons, and the nuance that the marker proves only a specific workflow is easily lost. If you’ve never used AI, you might still face suspicion because the tool doesn’t exonerate anyone. The marker’s limitations mean it’s simultaneously easy to evade and easy to misinterpret.

The real danger is chilling long-term. Fan fiction has always been a sandbox for experimentation—for new writers, multilingual authors, and creators working under pseudonyms. If the community’s default response to an imperfect provenance signal is a harassment campaign, those writers will simply stop publishing.

How AO3’s Policies Collided with Automated Vigilantism

Archive of Our Own’s official stance on AI is deliberately hands‑off—and for good reason. The Organization for Transformative Works (OTW), which runs the archive, does not ban AI‑assisted or AI‑generated fanworks. Its terms of service encourage creators to use relevant tags but explicitly state that the platform “cannot reliably enforce a universal disclosure rule.” The OTW has long argued that AI detection tools are unreliable and that removing works on suspicion alone would undermine the archive’s core mission: to preserve fan expression in all its variety, including imperfect or unpopular work.

The font-claude-response-body site skin complicates that stance. It’s not an official AO3 tool; it’s a third‑party skin that individual users can opt into. But because it automates an otherwise tedious inspection of page source, it acts as a de facto detection mechanism—one that AO3’s moderation team never authorized and has no ability to disable. The skin doesn’t remove works, but it primes readers to demand removal, and the resulting social pressure can accomplish what formal policy does not.

Critically, the marker itself is inspectable. A technically inclined reader can view a work’s page source, search for font-claude-response-body, and draw their own conclusions. The skin simply makes that inspection unavoidable. But automating a check is not the same as automating a fair judgment, and the archive’s governance was never designed to adjudicate disputes rooted in HTML remnants.

The Broader Pattern: AI Accusations Have Been Escalating

The Heated Rivalry dossier did not emerge in a vacuum. Fanlore, the community‑run documentation project for fandom history, has tracked a rising tide of AI‑related accusations and harassment throughout 2026. Its entry on the controversy notes that the flashpoints have included arguments over automated detection tools, public spreadsheets listing suspect stories, hostile comments, and authors putting works on hiatus as a result. The June 29 disclosure, with its veneer of technical objectivity, gave existing distrust a concrete vocabulary and a target list.

The dossier’s author, posting under the name John Doe, explicitly asked readers not to bully or ostracize the writers named within it. That plea was ignored. According to The New York Times, some authors faced waves of scrutiny and abusive comments; others locked their social media accounts, restricted access to their stories, or deleted entire bodies of work. The pattern is a familiar one online: a call for transparency morphs into a public trial, and the accused are expected to prove their innocence.

What You Should Do Right Now

The font-claude-response-body controversy isn’t going to vanish overnight, but there are practical steps you can take depending on your role in fandom.

If you’re a reader:
- Treat the site skin as a signal, not a verdict. If a page turns red, you’re seeing evidence of one possible workflow—not proof of deception.
- If you want to verify a work yourself, view the page source (right‑click and select “View Page Source”) and search for font-claude-response-body. The absence of the string doesn’t mean the work is AI‑free; its presence doesn’t tell you how the AI was used.
- Do not harass authors based on a marker alone. The dossier itself warned against bullying; adhere to that principle even when others don’t.

If you’re a writer:
- Audit your own paste habits. If you use Claude for drafting and want to minimize the risk of being flagged, always paste through a plain‑text intermediary.
- If you feel comfortable, consider adding an AI‑disclosure tag to your work. Be specific: “Brainstormed with Claude,” “Grammar checked with ChatGPT,” or “Partially AI‑generated” give readers a clearer picture than a binary label.
- If you’re targeted by a harassment campaign, you can moderate comments on AO3, restrict your works to registered users, or temporarily lock your social media accounts. Remember that you don’t owe the internet a full accounting of your creative process.

If you’re a community or platform leader:
- Advocate for better disclosure norms that distinguish between types of AI use. A single tag or CSS check cannot capture the nuance, but that doesn’t mean disclosure itself is worthless.
- Encourage platform‑level provenance features that travel with the text rather than relying on accidental formatting artifacts. Standards like the C2PA, which are being piloted for images, could eventually offer a more reliable trail.

Outlook: Provenance Tools Shouldn’t Be Improvised

The most unsettling part of the Heated Rivalry episode is how easily a technical quirk was repurposed into an accusation machine. Anthropic likely never intended font-claude-response-body to serve as a whistleblower. Yet it did—and without any built‑in context about what the marker can and cannot establish.

There is a clear product gap here. Readers crave trustworthy signals about a work’s origin. Writers need a way to disclose AI assistance without inviting exile. Platforms need tools that respect both nuance and user safety. Until those exist, communities will continue to weaponize whatever breadcrumbs they find, and the collateral damage will fall disproportionately on the very creators the archive was built to protect.

The red pages will fade from the Heated Rivalry tag, but the underlying tension will not. The next flashpoint will arrive as soon as someone spots another hidden marker—and the cycle will repeat until we learn to separate forensic facts from the moral panics they provoke.